← sources

Developing Krea 2 LoRA training for 12 GB GPUs — looking for feedback

source: reddit (reddit:StableDiffusion)published 2026-10-23 03:21ingested 2026-10-24 02:38

retained text

Hey everyone! I’m developing LoRA Studio, a local browser interface for preparing datasets, editing captions, and managing LoRA training with AI Toolkit. My current focus is Krea 2 training on 12 GB VRAM cards, with 8 GB as an eventual goal. Those are development targets; the tested development setup uses a 16 GB card.

I’d love for a few fellow enthusiasts to try the free technical preview and share feedback, especially on setup and memory use. No paid membership is required.

Download and instructions: [https://github.com/TigerFan69/lora-studio-preview](https://github.com/TigerFan69/lora-studio-preview)

Click “Download the application ZIP,” extract it, and follow docs/PREREQUISITES.md and docs/INSTALL.md inside. You’ll need Linux, an NVIDIA GPU, Python 3.12, an existing compatible AI Toolkit installation, and your own licensed model files. This is an add-on for an existing training setup; a clean-machine install is not yet verified.

The public download window runs through October 29 at 11:59 p.m. Central. Downloaded MIT-licensed copies remain usable afterward. Questions and feedback are welcome here or in the GitHub Issues tab!

Edit - my [Civit.ai](http://Civit.ai) page, well the one on red:  
[https://civitai.red/user/TigerFan69](https://civitai.red/user/TigerFan69)

document fetched 2026-10-24 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1xd31c7/developing_krea_2_lora_training_for_12_gb_gpus